Companies are increasing the number of ambitious announcements around artificial intelligence, but the uses are still struggling to become established in the daily lives of teams. To transform experiments into lasting results, leaders must now replace general promises with concrete, measurable and verifiable benefits.
“We place AI at the heart of our strategy. » “We are becoming an AI-native company. » “We are entering the era of intelligent agents. »
For two years, these declarations have multiplied in company communications. The tone is assured, resolutely turned towards the future. However, on the ground, the results often remain more modest than the ambitions displayed.
The latest figures published by INSEE, in July 2026, illustrate this discrepancy. In 2025, 18% of French companies with ten or more employees declared using at least one artificial intelligence technology, compared to 10% a year earlier. This rapid progression, however, masks a persistent obstacle: 54% of companies that do not use AI cite a lack of expertise. More than half of those who already use it say they encounter the same obstacle in the development of their uses (Insee, July 2026).
Internationally, Gartner estimates that more than 40% of agentic AI projects could be abandoned by the end of 2027, due in particular to increasing costs, poorly defined business value or insufficient controls (Gartner, June 2025).
There is therefore an obvious discrepancy. Leaders present artificial intelligence as a transformation comparable to the arrival of the Internet or electricity. At the same time, many employees are still trying to understand when, how and why they should use it.
This disconnect is usually described as an adoption problem. It is, more profoundly, a problem of trust.
Start from problems, not technology
Conversations about AI still too often start with the capabilities of models: multimodal reasoning, autonomous agents, computing power or size of language models.
However, these are not the subjects that concern users on a daily basis.
The questions asked by the teams are much more concrete: how to identify a project at risk before it falls into crisis? How can you avoid spending several hours each week manually producing progress reports? How to process and prioritize hundreds of requests without increasing staff numbers? How to quickly find reliable information dispersed between several tools?
It is from these irritants that any AI adoption strategy should begin.
A tool will not be used sustainably because it relies on impressive technology. It will be because it allows a person to accomplish a task more easily, more quickly or with more precision.
Users are not necessarily looking for a spectacular assistant, capable of impressing during a demonstration but which becomes useless as soon as the work becomes more complex. They want a tool that understands their environment, reduces non-value-added tasks, and provides answers they can actually rely on.
The lack of expertise highlighted by INSEE must also be interpreted beyond just the shortage of technical skills. It also reveals the difficulty organizations face in linking the possibilities of AI to specific business processes, in supervising uses and in defining the criteria for evaluating the results.
Replace promises with proof
Telling an organization that it should “adopt AI” is almost as abstract as advising it to “go digital more.”
To gain the trust of teams such as financial departments, companies must become much more precise. The right unit of measurement is not only the number of licenses deployed, the number of employees trained or the volume of content generated. This is the use case and the result it produces.
In the field, the most effective approach is to start by identifying repetitive tasks and the main friction points in operational processes. Each problem is then associated with a specific AI capability. This work makes it possible to generate measurable time savings before considering deployment in other functions.
This method is less spectacular than a global transformation plan, but it quickly provides an answer to the essential questions: is the tool really used? Does it improve the quality of work? Does it free up time? Does it reduce delays or errors?
The most revealing indicators do not always require a complex measuring device. They can take the form of shorter meetings, accelerated validation cycles, simplified information searches or faster deliveries.
Certain international digital agencies, by integrating modern project management solutions like Wrike (Jellyfish case study)thus noted savings of three to five hours per person per week thanks to AI, in particular through a drastic reduction of 95% in the time spent summarizing discussions and actions decided during customer calls.
In the legal advisory and business services industry, centralizing work, automating reporting and improving project visibility can cut the time spent on follow-up meetings in half, while saving around 20% in time by avoiding manual document searching.
These gains may appear modest when viewed in isolation. But when they concern several hundred employees and are repeated every week, their impact becomes considerable.
They are also indicators that everyone can understand. They stand up to budget scrutiny and enable concrete benchmarks to be established for future deployments.
Giving AI business context
Most AI tools don’t produce unsatisfactory answers because their underlying technology is bad. They fail because they don’t know enough about the organization they’re supposed to help.
A general model can explain how to build a project plan. But can it identify which projects are actually behind schedule in a given company? Does he know his business priorities, the responsibilities of each team, the dependencies between tasks and the decisions made in the last meetings?
Without access to this context, AI necessarily produces generic responses.
To become truly useful, it must be able to rely on the company’s operational data, understand its processes and respect the access rights associated with each piece of information. This issue is particularly important in France and Europe: organizations cannot dissociate the adoption of AI from issues of confidentiality, data governance, cybersecurity and regulatory compliance.
Connecting an AI to the business context does not mean giving it indiscriminate access to all information. This involves building a structured and controlled knowledge environment, in which each user can only use the data to which they are authorized to access.
The AI must also be able to indicate what information it relies on. The more an answer influences an important decision, the more its origin must be understood and verified by the user.
It is this combination of context, traceability and control of permissions that can move AI from the status of an experimental tool to that of a sustainable component of operations.
Make trust a performance indicator
Trust in AI cannot be decreed through a communications plan. It is built by accumulation of positive experiences.
When a tool provides an exact answer, saves twenty minutes on a tedious task or helps anticipate a problem, it slightly reinforces this confidence. Conversely, each answer that is wrong, disconnected from context, or impossible to verify creates an additional reason to revert to the old ways.
Adoption therefore depends as much on the quality of these daily interactions as on the major strategic orientations.
It also means giving teams the means to understand the tools, monitor their results and report errors. Trustworthy AI is not a black box to which a decision is blindly delegated. It is a system that assists the user while allowing them to maintain their ability to judge.
Training must also evolve. A general presentation of the possibilities of AI is not enough. Employees need to learn how to use it in their own profession: to formulate a useful request, to identify relevant data, to verify a response and to recognize situations in which human intervention remains essential.
The organizations that progress the fastest generally follow a simple approach: identify a specific friction point, select an appropriate response, measure the result, collect user feedback, then gradually expand the scope.
Moving from “AI-first” ambition to demonstrated utility
Perhaps leaders today who say they want to become “AI-first” should start by answering two simpler questions: Where is this technology already working in our organization? And how does it actually help its users?
If the response requires lengthy reservations, a warning that this is still only a pilot, or a reference to a future roadmap, the credibility gap remains open.
The challenge for 2026 is no longer to convince companies that artificial intelligence will have a major impact. This conviction is now widely shared.
The challenge is to transform this ambition into visible, measurable and repeatable improvements. To give tools the context they need. To evaluate their value on the scale of real uses. And to fully involve employees in their deployment.
Ultimately, the success of AI in business will not depend solely on the power of the models. It will be based on a profoundly human quality: trust.
Thomas Scott, CEO of Wrike and Klaxoon by Wrike
